Treatment Heterogeneity in Comparative Effectiveness of Teriparatide vs. Bisphosphonates: Multi-database Cohort Study

Background: Osteoporosis is associated with increased fracture risk, affecting almost 30% of women aged >50 years. Bisphosphonates (BP) and teriparatide (TP) both reduced hip fracture risk in clinical trials, but large-scale studies evaluating their comparative effectiveness in real-world settings are lacking.

Objectives: To study the comparative effectiveness of TP vs BP to reduce hip, major osteoporotic and vertebral fracture risk. In addition, we stratified by predicted hip fracture risk to assess treatment effect heterogeneity.

Methods: We conducted a network cohort study using data from IBM MarketScan® Commercial Claims and Encounters (CCAE), Optum® De-Identified Clinformatics Data Mart Database — Date of Death (OPTUM-DOD) and Optum® de-identified Electronic Health Record Dataset (OPTUM-EHR), all mapped to the OMOP common data model.

We included all women aged >50, who initiated TP or BP and had no history of anti-osteoporotic treatment in the prior year. Propensity scores were used for 1:10 matching to minimise confounding by indication.

We conducted overall and risk stratified analyses. Models to predict hip fracture risk were developed and validated in all 3 databases.

We checked covariate imbalance excluding analyses with residual imbalance after matching.

Finally, 147 negative control outcomes were included to calibrate for residual confounding. Cox regression was used to estimate calibrated hazard ratios (HR) and Kaplan-Meier estimated differences on day 365 to estimate absolute effects. We provide meta-analytic estimates for the overall analysis and for the bottom 75% and top 25% hip fracture risk subpopulations.

Results: A total of 8,691 and 307,778 users of TP and BP were included from CCAE, 9,532 and 514,926 from Optum-DOD, and 10,164 and 197,250 from OPTUM-EHR. Overall meta-analytic calibrated HR were 0.92 [0.76-1.12], 1.17 [1.00-1.36] and 1.20 [1.00-1.43] for hip fracture, major osteoporotic fracture and vertebral fracture respectively. Meta-analytic calibrated HRs for the lower 75% hip fracture risk subpopulation were 0.75 [0.52-1.08], 1.31 [1.11-1.54], and 1.42 [1.12-1.79], respectively. In the top 25% risk subpopulation the respective estimates were 0.98 [0.75-1.29], 1.10 [0.94-1.29], and 1.04 [0.85-1.26]. Our negative control analyses showed evidence of residual confounding.

Conclusions: Our study found relevant treatment heterogeneity, with a tendency towards favouring TP in patients with low anticipated hip fracture risk. However, negative control outcome analyses pointed towards unresolved confounding.